gowalk-cicd 1.0.195 → 1.0.197

package/backend-action/health_rollout.py

@@ -50,7 +50,7 @@ def candidate(engine, old, body, name, staged, reservation):
     current = engine.inspect(old['Id'])
     unchanged(old, current)
     if current['State']['Running']:
-        engine.request('POST', '/containers/' + old['Id'] + '/stop?t=10')
+        engine.stop(old['Id'])
     replacement = engine.inspect(name, missing=True)
     if replacement is None:
         result = engine.request('POST', '/containers/create?name=' + quote(name, safe=''), body)
@@ -71,6 +71,8 @@ def confirmed(engine, method, path, proof):
 
 
 def rollback(engine, old, name, reservation):
+    if engine.pending_stop is not None:
+        engine.stop(old['Id'])
     replacement = engine.inspect(name, missing=True)
     if replacement is not None and replacement['Id'] != old['Id']:
         if (replacement['Config'].get('Labels', {}).get(MARKER) != old['Id']

package/backend-action/health_rollout_engine.py

@@ -13,14 +13,15 @@ SOCKET = '/var/run/docker.sock'
 class Engine:
     def __init__(self):
         self.deadline = time.monotonic() + 90
+        self.pending_stop = None
 
-    def request(self, method, path, body=None, missing=False, raw=False):
+    def request(self, method, path, body=None, missing=False, raw=False, timeout=10):
         remaining = self.deadline - time.monotonic()
         if remaining <= 0:
             raise ValueError('rollout_deadline')
-        connection = http.client.HTTPConnection('localhost', timeout=min(10, remaining))
+        connection = http.client.HTTPConnection('localhost', timeout=min(timeout, remaining))
         connection.sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
-        connection.sock.settimeout(min(10, remaining))
+        connection.sock.settimeout(min(timeout, remaining))
         try:
             connection.sock.connect(SOCKET)
             data = None if body is None else json.dumps(body).encode()
@@ -38,6 +39,15 @@ class Engine:
     def inspect(self, identifier, missing=False):
         return self.request('GET', '/containers/' + quote(identifier, safe='') + '/json', missing=missing)
 
+    def stop(self, identifier):
+        # Docker continues an accepted stop after a lost HTTP reply. Keep that uncertainty
+        # until the same-ID stop completes and its stopped state is positively observed.
+        self.pending_stop = identifier
+        self.request('POST', '/containers/' + identifier + '/stop?t=10', timeout=20)
+        if self.inspect(identifier)['State']['Running']:
+            raise ValueError('container_stop_unconfirmed')
+        self.pending_stop = None
+
 
 def unshared(engine, original):
     identifiers = [row['Id'] for row in engine.request('GET', '/containers/json?all=1')]
@@ -67,9 +77,16 @@ def probe(engine, identifier, test):
     created = engine.request('POST', '/containers/' + identifier + '/exec',
                              {'AttachStdout': False, 'AttachStderr': False, 'Cmd': command})
     engine.request('POST', '/exec/' + created['Id'] + '/start', {'Detach': False, 'Tty': False}, raw=True)
-    result = engine.request('GET', '/exec/' + created['Id'] + '/json')
-    if result['Running'] or result['ExitCode'] != 0:
-        raise ValueError('original_probe_failed')
+    deadline = min(time.monotonic() + 5, engine.deadline)
+    while time.monotonic() < deadline:
+        result = engine.request('GET', '/exec/' + created['Id'] + '/json',
+                                timeout=max(0.001, deadline - time.monotonic()))
+        if not result['Running']:
+            if result['ExitCode'] != 0:
+                raise ValueError('original_probe_failed')
+            return
+        time.sleep(min(0.1, max(0, deadline - time.monotonic())))
+    raise ValueError('original_probe_timeout')
 
 
 def healthy(engine, identifier, deadline):

package/backend-action/health_rollout_spec.py

@@ -113,7 +113,12 @@ def unchanged(before, after, reservation_changes=False):
 
 def verified(row, body, original):
     config = {key: value for key, value in body.items() if key not in ('HostConfig', 'NetworkingConfig')}
-    if (row['Config'] != config or row['Image'] != original['Image'] or row['HostConfig'] != body['HostConfig']):
+    hosts = [dict(value) for value in (row['HostConfig'], body['HostConfig'])]
+    # Docker reports the default OOM setting as null while running and false before start.
+    for host in hosts:
+        if host.get('OomKillDisable') is None:
+            host['OomKillDisable'] = False
+    if row['Config'] != config or row['Image'] != original['Image'] or hosts[0] != hosts[1]:
         raise ValueError('replacement_config_changed')
     actual = row['NetworkSettings']['Networks']
     expected = body['NetworkingConfig']['EndpointsConfig']

package/backend-action/runtime_image_health.py

@@ -5,6 +5,7 @@ import re
 import subprocess
 
 from runtime_healthcheck import bounded_tmpfs, cache_healthcheck
+from runtime_model import escape_interpolation
 
 
 def eligible(service):
@@ -17,7 +18,14 @@ def eligible(service):
 def image_health(command, name, service):
     reference = service.get('image')
     if not reference:
-        result = subprocess.run(command + ['config', '--images', name],
+        resolver = list(command)
+        option = next((flag for flag in ('--file', '-f') if flag in resolver), None)
+        if option:
+            resolver[resolver.index(option) + 1] = '-'
+        else:
+            resolver.extend(['--file', '-'])
+        selected = {'services': {name: {'build': '.'}}}
+        result = subprocess.run(resolver + ['config', '--images', name], input=json.dumps(selected),
                                 capture_output=True, text=True, timeout=30, check=True)
         references = result.stdout.split()
         if len(references) != 1:
@@ -44,6 +52,7 @@ def inherit_healthchecks(model, command):
         candidate = copy.deepcopy(service)
         candidate.setdefault('healthcheck', {})['test'] = health['Test']
         if cache_healthcheck(candidate):
+            candidate['healthcheck']['test'] = escape_interpolation(candidate['healthcheck']['test'])
             service['healthcheck'] = candidate['healthcheck']
             service.update(image=image['Id'], pull_policy='never')
             changed = True

package/docs/BACKEND_HEALTHCHECKS.md

@@ -36,6 +36,10 @@ wraps only a missing Compose test, retaining explicit timing and disabled/custom
 launch model pins the transformed service to that inspected image ID with pulling disabled, so a
 moving tag cannot substitute another image's healthcheck. Explicit registry-refresh policies and
 non-build services retain their existing image/pull behavior. No pre-build image healthcheck is reused.
+Imported image commands escape literal dollars once at the Compose boundary, preserving Python
+strings and container-shell expansion without interpreting them against the deployment host environment.
+Default image names resolve from an isolated service model with the same Compose project name;
+dependency images cannot make that selection ambiguous.
 
 ## Adopt an existing API
 
@@ -81,6 +85,12 @@ marker-bound replacement, restore/start the original and verify its original hea
 by identity/state readback. `rollback_unconfirmed` means recovery needs another exact readback/replay;
 retained containers and records stay available, and unrelated containers are never removed to clear a name.
 
+Docker may report the default `OomKillDisable` as either null or false; these alone compare equivalently,
+while an enabled OOM-kill override must match exactly. Stop requests retain their ten-second grace with
+a twenty-second transport allowance inside the transaction budget. An uncertain stop must settle on
+that exact container before rollback can restart it or certify its health. Original-probe verification
+waits for the exec's confirmed exit status, for at most five seconds within the remaining transaction.
+
 Replay with the same original ID resumes an interrupted retained-container transaction. After successful
 cleanup it finds the marker-bound healthy replacement and returns `already_applied`. It refuses absent,
 ambiguous or changed runtime evidence. Start with one canary, verify application health externally and

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "gowalk-cicd",
-  "version": "1.0.195",
+  "version": "1.0.197",
   "description": "Zero-config GitHub Actions delivery for iOS TestFlight and Android Google Play.",
   "type": "module",
   "bin": {